feat(model): wählbares LLM-Modell pro User & pro Job

Mac-Worker:
- /api/summarize, /api/protocol, /api/preload akzeptieren optionales
  model-Feld → überschreibt den .env-Default pro Call, kein Restart nötig
- /api/models listet die lokal verfügbaren Ollama-Modelle + Default

LXC:
- User.default_model + Job.model (Migration für SQLite)
- /api/auth/me liefert default_model, PATCH speichert ihn sofort
- /api/mac/models proxiet die Liste an die UI
- Pipeline reicht job.model an Mac-Worker durch; leer = Worker-Default
- create_job & reprocess akzeptieren model-Override
- Snapshot-Meta nutzt job.model (genauer als /health-Snapshot)

UI:
- Header-Dropdown "Modell:" neben "Profil:" — sofort-speichert User-Default
- Jeder Job zeigt sein Modell als Tag (lila); leer bedeutet Worker-Default
- Reprocess öffnet einen Modal-Picker (statt simplem confirm), Auswahl
  aller verfügbaren Modelle inkl. "Worker-Default"

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

lxc-frontend/app/services/pipeline.py

@@ -97,6 +97,8 @@ async def run_pipeline(job_id: int) -> None:
         audio_path = settings.upload_dir / job.filename
         title = job.title or audio_path.stem
         profile = job.profile or "meeting"
+        # Leeres `model` → Mac-Worker nutzt seinen .env-Default.
+        model = (job.model or "").strip() or None
         existing_tx_path  = _existing(job.transcript_path)
         existing_sum_path = _existing(job.summary_path)
         existing_pro_path = _existing(job.protocol_path)
@@ -131,16 +133,17 @@ async def run_pipeline(job_id: int) -> None:
             _update(job_id, status=JobStatus.SUMMARIZING, progress=70)
         else:
             _start_step(job_id, JobStatus.SUMMARIZING, 50)
-            log.info("[job %s] preload model", job_id)
+            log.info("[job %s] preload model=%s", job_id, model or "(worker default)")
             try:
-                await _with_heartbeat(job_id, client.preload(profile=profile, job_id=job_id))
+                await _with_heartbeat(job_id, client.preload(profile=profile, model=model, job_id=job_id))
             except Exception as e:  # noqa: BLE001
                 log.warning("[job %s] preload failed (non-fatal): %s", job_id, e)
             _update(job_id, progress=55)
             log.info("[job %s] summarize", job_id)
             summary = await _with_heartbeat(
                 job_id,
-                client.summarize(transcript_text, title=title, profile=profile, job_id=job_id),
+                client.summarize(transcript_text, title=title, profile=profile, model=model,
+                                 job_id=job_id),
             )
             summary_path = job_dir / "summary.json"
             summary_path.write_text(json.dumps(summary, ensure_ascii=False, indent=2), encoding="utf-8")
@@ -156,7 +159,8 @@ async def run_pipeline(job_id: int) -> None:
             log.info("[job %s] protocol", job_id)
             protocol = await _with_heartbeat(
                 job_id,
-                client.protocol(transcript_text, summary, title=title, profile=profile, job_id=job_id),
+                client.protocol(transcript_text, summary, title=title, profile=profile, model=model,
+                                job_id=job_id),
             )
             protocol_path = job_dir / "protocol.json"
             protocol_path.write_text(json.dumps(protocol, ensure_ascii=False, indent=2), encoding="utf-8")